Have you considered using an annotation processor instead of byte code enhancement?
Have you considered transforming the source code prior to compile instead of byte code enhancement?
Another option is to introduce another subproject to transform the jar and depend on that instead of the original.